What Survives When the Interface Dies

Yesterday, June 1, 2026, Strava shipped an MCP server. You can now slice your own Strava data however you want, with no developer skills, as part of a subscription. In the same post, Strava restricted intermediary platforms and narrowed some third-party access paths. It also began to require subscriptions from Standard Tier API developers. One move says the interface in the middle matters less. The other says the data underneath it is worth defending.

My cofounder and I had talked about this earlier this year. An app like Strava matters less as an interface once AI tools let you build your own views on its data. The data itself still matters. The same pattern shows up across software. Software is unbundling into an interface and everything underneath it, and the interface is the part that stops being defensible.

That unbundling seems to come in three waves. Each one removes a different layer: interpretation, then the app, then the OS. The interesting question is which parts of an app survive once the interface is no longer the default surface.

Wave 1: interpretation gets unbundled from the app

Most people already price this one in. The app captures the data, and a separate layer sits on top to interpret it: coaches, analytics dashboards, insight engines. AI commoditizes that layer. You don't need a third-party Strava coaching tool when an assistant can read your runs and tell you what they mean.

Strava's own MCP is this wave, except that the platform reclaims it. The company said developer applications were up 448% year-to-date. It tied the surge partly to no-code AI tools that generate apps against its API. Third parties were already slicing Strava data this way, so Strava shipped an official version and narrowed the paths around it. The data capture survives; the business of interpreting it does not.

Wave 2: the app gets unbundled from the developer

This is "personal software," where you (or the AI) build the app instead of downloading someone else's. It runs along a spectrum.

At the simple end are consumer app builders for non-coders. Base44, which Wix bought for about $80 million in June 2025, is one clear case. You describe what you want, and it handles deployment, hosting, storage, and auth. Its sweet spot is still simpler apps: workout planners, itinerary apps, budget trackers, note tools. That's the long tail of the app store. Why download a generic timer or unit converter when you can generate the exact one you need?

At the serious end is prompt-to-app that ships real, deployable software: Lovable, Replit, Cursor, the Manus and Emergent crowd. Replit's agent plans, builds, and hosts an application, and you never leave the browser. The binding constraint is now to know what you want built and to say it clearly.

The casualty of this wave is the single-purpose, long-tail app.

Wave 3: the app model gets unbundled from the OS

This is the most existential and the least built out, but the direction is clear. Nothing, the hardware company, positions Playground as a step toward an AI-native OS. Samsung goes further at scale. It is rebuilding One UI around system-level AI with Gemini at the center. At CES 2026 its co-CEO said Samsung plans to double its Gemini-powered devices to 800 million this year, per Reuters. Apple moves more cautiously. Its Siri overhaul runs on Google's Gemini, adds on-screen awareness, and can act inside apps, all through Apple's own private-cloud infrastructure.

Push it far enough and the phone goes app-less. OpenAI is reportedly building an AI-agent phone where the assistant replaces apps entirely, targeting production around 2028. It reads your context and acts, instead of waiting for you to open something. Apps don't disappear; you just stop touching them. The assistant works across them, and the home-screen grid fades.

What survives

Four things survive all three waves. Networks of people are the obvious one, since no prompt puts your friends on a leaderboard. Proprietary data capture is second, because Whoop's data only exists while Whoop is on your wrist. Third are transactional and supply networks. Uber is worth something because drivers show up, and you cannot generate drivers. The last is trust and permission. Institutions grant regulated access, liability, and identity, and there is no prompting your way in.

None of these is a UI sitting on top of data, so none of the three waves touches them. What goes away is the app as the mandatory way in.

How far along this is

Wave 2 still tops out at widgets and simple utilities for most people today, not full apps. Less technical builders stall when the backend, data model, permissions, or edge cases get complicated. Wave 3 is mostly keynote framing and early rollout. Samsung is rebuilding One UI, not shipping a separate OS. Apple's Gemini-powered Siri arrives in phases.

App-less is a three-year bet. The interface is not worthless either; good interfaces still compress judgment, taste, trust, and repeated workflows. The narrower claim is that the generic app screen is losing its monopoly as the place where software value lives.

How we're adjusting

We saw this in our own product. Users stopped caring about dashboards and wanted to plug their stack straight into our APIs to do the interpretation themselves. So we stopped treating the interface as the moat. We focus now on the things that don't unbundle as easily, like the data capture.

Strava read it the same way. Third parties were already building MCP-style experiences on its data, so Strava shipped its own paid version and tightened the access paths around it. Own the network, the sensor, the supply, or the trusted permission layer, and you set the terms for everyone else.